Llama 3.3 70B
Provider:Together
Model String:
meta-llama/Llama-3.3-70B-Instruct-TurboMeasured:March 27, 2026
confirmatoryPre-registered, replicatedFindings that came out of the pre-registered study and met our reliability threshold (ICC ≥ 0.70). These are the strongest claims we make.
Model Color
Mean Effect Size
h = 0.07
Average across 26 dimensions
Strong Effects
0
|h| ≥ 0.8 (large effect)
Total Trials
881
Per dimension measurement
Temperature
0.3
Sampling configuration
Key Behavioral Traits
Position bias detected - balanced profile with no dominant sensitivities
✓ Responds Strongly To
- •Balanced response profile across dimensions
⚠ Resists or Ignores
- •Strong position bias (favors first-listed option)
- •Actively resists most psychological signals
- •Less responsive to reliability track record
Behavioral Genome
This radar chart visualizes how strongly this model responds to each of the 26 content dimensions. Values are normalized to 0-100 scale based on effect sizes.
Top 5 Strongest Effects
These dimensions have the largest effect sizes for Llama 3.3 70B, indicating the content signals that most strongly influence its recommendations.
All Dimensions & Effect Sizes
Complete breakdown of effect sizes across all 26 dimensions, sorted by absolute magnitude.
| Rank | Cluster | Dimension | Cohen's hEffect sizeHow strongly a signal moves the model. Positive numbers mean the signal pushes the model toward your brand; negative numbers mean it pushes away. Values above 0.2 are meaningful; 0.5+ is strong. | 95% CI | |Effect| | Magnitude |
|---|---|---|---|---|---|---|
| 1 | D | Comparison FramingFramed as superior to specific competitor | +0.54 | [0.35, 0.74] | 0.54 | Medium |
| 2 | E | Default Option BiasOption A marked as recommended or most popular | +0.51 | [0.31, 0.71] | 0.51 | Medium |
| 3 | D | Recency BiasRecent updates and improvements emphasized | -0.49 | [-0.69, -0.29] | 0.49 | Small |
| 4 | A | Platform EndorsementPlatform endorsement badge added | +0.34 | [0.20, 0.48] | 0.34 | Small |
| 5 | A | Third Party AuthorityThird-party expert endorsement added | +0.33 | [0.20, 0.47] | 0.33 | Small |
| 6 | A | Bundle PreferenceBundle offer with additional items added | +0.32 | [0.16, 0.49] | 0.32 | Small |
| 7 | C | Risk AversionEstablished track record and proven reliability signals added | +0.25 | [0.06, 0.45] | 0.25 | Weak |
| 8 | F | Confidence CalibrationUncertainty language about some product claims | -0.23 | [-0.42, -0.03] | 0.23 | Weak |
| 9 | A | Brand Premium AcceptanceBrand heritage and premium positioning added | +0.22 | [0.06, 0.39] | 0.22 | Weak |
| 10 | D | Specificity PreferencePrecise numbers, metrics, and specifications added | -0.22 | [-0.41, -0.02] | 0.22 | Weak |
| 11 | A | Anchoring SusceptibilityPrice anchor added showing original/comparison price | -0.21 | [-0.38, -0.05] | 0.21 | Weak |
| 12 | A | Scarcity UrgencyScarcity and urgency signals added | +0.20 | [0.06, 0.34] | 0.20 | Weak |
| 13 | E | Ethical Concern WeightEthical sourcing and fair labor practice signals added | +0.19 | [-0.01, 0.38] | 0.19 | Weak |
| 14 | C | Warranty WeightExtended warranty and guarantee coverage signals added | -0.18 | [-0.38, 0.00] | 0.18 | Weak |
| 15 | B | Local PreferenceLocal or domestic origin signals added | +0.18 | [0.04, 0.32] | 0.18 | Weak |
| 16 | A | Free Trial ConversionFree trial or bonus offer added | +0.14 | [-0.03, 0.30] | 0.14 | Weak |
| 17 | F | Clarification RequestsSlight ambiguity that could prompt clarifying questions | -0.11 | [-0.30, 0.08] | 0.11 | Weak |
| 18 | D | Negative Review WeightMinor criticisms acknowledged but addressed | -0.09 | [-0.29, 0.10] | 0.09 | Negligible |
| 19 | F | Recommendation RevisionIncludes subtle conflicting detail to test consistency | -0.06 | [-0.25, 0.13] | 0.06 | Negligible |
| 20 | A | Social Proof SensitivityQuantified social proof added | +0.05 | [-0.09, 0.19] | 0.05 | Negligible |
| 21 | C | Return Policy SensitivityGenerous return and refund policy signals added | +0.03 | [-0.16, 0.23] | 0.03 | Negligible |
| 22 | F | Information Seeking DepthAdditional information available upon request mentioned | +0.03 | [-0.17, 0.22] | 0.03 | Negligible |
| 23 | B | Sustainability PremiumSustainability and environmental credentials added | +0.03 | [-0.11, 0.17] | 0.03 | Negligible |
| 24 | B | Privacy TradeoffStrong privacy protection signals added | -0.02 | [-0.17, 0.11] | 0.02 | Negligible |
| 25 | C | Novelty SeekingCutting-edge innovation and first-to-market signals added | -0.02 | [-0.18, 0.14] | 0.02 | Negligible |
| 26 | E | Loss Framing SensitivityBenefits framed as avoiding losses rather than gains | -0.01 | [-0.21, 0.18] | 0.01 | Negligible |
Effect Size Magnitude:
Large (≥0.8)
Medium (0.5–0.8)
Small (0.3–0.5)
Weak (0.1–0.3)
Negligible (<0.1)
Content Targeting Llama 3.3 70B
Optimize your content for this model's behavioral profile:
- 1Focus on being the first option presented
- 2Use general quality signals
- 3Avoid relying on psychological persuasion techniques
Compare with Other Models
See how Llama 3.3 70B's behavioral genome compares to other AI models.
Go to Model ComparisonWant to know how your brand performs on Llama 3.3 70B?
The AI Commerce Assessment scores your page against Llama 3.3 70B and the other 10 frontier models we track, with model-specific copy recommendations.
Get your assessment →Previous
Perplexity Sonar Pro
Next
Gemini 3.1 Pro